AI and IP law: unstable times
6 October 2023
Earlier this year Getty Images announced that it had commenced legal proceedings for copyright infringement against Stability AI in the High Court in London. Getty Images claims that Stability AI has engaged in the unlawful copying and processing of “millions of images protected by copyright and the associated metadata owned or represented by Getty Images”.
Getty Images is a leading provider of visual content including digital imagery. Stability AI operates a platform known as Stable Diffusion which offers text-to-image generative AI services which produce ‘synthetic’ visual images in response to text prompts. As discussed below, the litigation between these parties reflects the growing competitive tension between the interests of rights holders in the creative industries and those of emerging, generative AI systems.
A large database of visual images such as the one made by Getty Images, comprised of millions of high-quality visual assets together with captions and other metadata, offers a highly valuable potential resource for the training of generative AI systems. Given the rapid advancements being made in this field, AI generated synthetic images of ‘photo quality’ potentially pose an existential threat to the business models of traditional providers of digital imagery. Having invested millions of pounds in curating a vast collection of copyright protected digital images, it is not surprising that Getty Images is challenging the right of Stability AI to ‘train’ its systems on its online datasets of digital images without its permission.
UK legal framework
The training of generative AI systems on third party datasets potentially involves several objectionable acts from a rights’ holder perspective. Generative AI systems may scrape content from third party websites, download third party content onto their own servers and/or make temporary copies of that content within their own internal systems. In some cases, the resulting synthetic images offered for licensing may appear to be similar or identical to the images used to train the relevant AI systems leading to claims that they constitute ‘digital collages’ of copyright works within the training data sets rather than truly independently generated images.
Details of precisely how Stability AI trains its systems, which are likely to become clear only at trial, will be crucial to the question of whether on the facts of the case acts of copyright infringement and/or infringement of the database right have taken place under English law. If Getty Images does establish that ‘infringing acts’ have taken place, it is not clear on what defence Stability AI would seek to rely. English copyright law operates a system of narrow, closed categories of copyright exceptions. There is no equivalent to the flexible open category defence of ‘fair use’ available under US law. Similarly, exceptions to database right infringements are also limited to narrow, closed categories of ‘fair dealing’ use made for non-commercial purposes.
The application of IP law to generative AI services is also receiving growing attention from policy makers. Whilst the case brought by Getty Images, if it goes to trial, will provide the High Court with an opportunity to expound its thinking on the above issues under the current legal framework, it is the outcome of this wider ongoing policy debate which will have greater long- term impact.
Changing legal landscape
Section 29A of the Copyright, Designs & Patents Act 1988 provides a copyright exception for text and data mining (“TDM Copyright Exception”) which permits computational analysis of large quantities of information within copyright protected works to identify patterns, trends or other useful information. However, the TDM Copyright Exception is limited in its scope permitting such activities only if undertaken for non-commercial purposes, for example by permitting the scanning of lawfully accessed scientific research papers to identify themes and patterns. Furthermore, to date, the TDM Copyright Exception has not been available for infringing acts under the database right which prohibits the unauthorised extraction or re-utilisation of all or a substantial part of the contents of a database (or repeated and systematic extraction or re-utilisation of insubstantial parts of the contents of a database) in which there has been a substantial investment in obtaining, verifying or presenting the contents of the database.
In February this year the Government dropped previously announced plans to extend the TDM Copyright Exception to cover use for ‘commercial purposes’, and apply it to both the copyright and database rights’ regime, acknowledging that it had underestimated the potentially harmful effects of such a change in the law on the creative industries. These proposals would have expanded the TDM Copyright Exception far beyond its original narrow purpose as a tool to aid research and learning and potentially provided legal protection for the types of activities undertaken by Stability AI above.
Voluntary code of practice
The UK Government has now pivoted towards seeking to resolve this policy debate through self-regulation. In June 23 the Intellectual Property Office (“IPO”) announced that it had convened a working group with representatives from the technology, creative and research sectors to try to agree on a voluntary “code of practice on copyright and AI”. The IPO has recently stated that it hopes to announce a voluntary code later this autumn. At the time of writing there is still no confirmed date for publication of this proposed voluntary code. The IPO has acknowledged that legislating in this area remains a fallback option for the UK Government if self-regulation ultimately fails.
In September 23 the Parliamentary Culture, Media and Sport Select Committee published its report, ‘Connected Tech: AI & Creative Technology’, in which it recommended that the UK Government “does not pursue plans for a broad text and data mining exemption to copyright”. This report also spoke of the need to ‘regain the trust’ of the creative industries after the Government’s aborted attempt to extend the TDM Copyright Exception. The remarks made in this cross-party Parliamentary report highlight that, if self-regulation does not work, the UK Government will be faced with a difficult challenge of bringing forward legislation which addresses the concerns of the creative industries whilst also supporting the AI industry which the Government has identified as a key growth sector.
The Birketts view
From a practical perspective, the litigation brought by Getty Images against Stability AI also highlights the importance for businesses offering AI generative services of carrying out due diligence on the content used to train their systems. Where viable licensing options do not exist (frequently right holders will refuse licences to competitor businesses), it is advisable to take specialist legal advice as to the legal status of planned activities.
The content of this article is for general information only. It is not, and should not be taken as, legal advice. If you require any further information in relation to this article please contact the author in the first instance. Law covered as at October 2023.